Transcript Week3 - Cat`s TCM Notes

Evidence Based Medicine
Week 3:
Basic Research Concepts in Western and
Eastern Medicine.
Part II: Statistics
Terminology Review...



Variable (things we measure)
Correlational vs. Experimental research
(experiment has manipulation and control
and answers “Why” or “how” questions.
Correlational just shows how things are or
were. No manipulation, just observation.)
Dependent vs. Independent variable. (You
manipulate the independent variable, and
the dependent variable reacts.)
Measurement scales for
variables




Nominal – category like race or gender.
Ordinal – this gives rank. We can know
something is the best, but that doesn't tell us
anything about how much better it is than other
things.
Interval – these give us rank and also a reliable
quantity - quantify the variables. Like
temperature. 30 degrees is 10 more than 20...
Ratio variables -rank, quantity and a solid zero
point so mathematically more precise. Like
temperature in degrees Kelvin or time.
Relationships between variables



Ultimately, every study examines the
relationships between variables.
The whole point is finding out how one
variable changes the other (i.e. treatment
and pain/health. Diet and obesity,
acupuncture and depression, whatever...)
Statistics gives us a tool to evaluate the
strength of the relationship that we find,
and also the probability that whatever we
find just happened by chance.
Statistical significance






“p” value represents the probability that the
results of the study happened by random
chance.
This is the degree to which the result can be
considered “true” “useful” or “representative” of
the general population.
p=0.05 means there's a 5% chance of the result
being a fluke.
The lower the “p” value the more valid the result.
So p=.001 is much better than p=.05
P=.05 is the minimum standard for a result to be
considered “significant.”
“p” measures “reliability” or “truthfulness”
Magnitude/Size/Strength of
Relationships





The stronger the relationship between variables
found in a study, the more likely it is that the
relationship also exists in the general population.
“strength” of relationship and “reliability” are
therefore related.
This is only true if the sample size is kept
constant – see the next slide.
Sample size is represented by “n” so n=150
means 150 people participated in the study.
N=1000 means 1000 people.
In general bigger “n” means better results.
Size of the Sample


If you only measure one person, then it's
easy to find relationships between things
that aren't actually related or
representative of the general population.
Like saying because one woman (gender)
is blond (hair color), thus all women are
blond.
The more people you measure, the more
likely it is that whatever trends or
relationships you see between variables
are actually true.
Example

Consider this example from research on statistical
reasoning (Nisbett, et al., 1987). There are two
hospitals: in the first one, 120 babies are born every
day; in the other, only 12. On average, the ratio of
baby boys to baby girls born every day in each
hospital is 50/50. However, one day, in one of those
hospitals, twice as many baby girls were born as
baby boys. In which hospital was it more likely to
happen? The answer is obvious for a statistician, but
as research shows, not so obvious for a lay person:
it is much more likely to happen in the small
hospital. The reason for this is that technically
speaking, the probability of a random deviation of a
particular size (from the population mean),
decreases with the increase in the sample size.
Smaller relationships need larger
samples

If a coin is slightly asymmetrical and, when tossed,
is more likely to produce heads than tails (e.g., 60%
vs. 40%), then ten tosses would not be sufficient to
convince anyone that the coin is asymmetrical.
However, if the effect in question were large enough,
then ten tosses could be enough. For instance,
imagine that the coin is so asymmetrical that no
matter how you toss it, the outcome will be heads. If
ten tosses produced ten heads, most people would
consider it significant. In other words, it would be
considered convincing evidence that in the
theoretical population of an infinite number of tosses
of this coin, there would be more heads than tails.
Thus, if a relation is large, then it can be found to be
significant even in a small sample.
“No Relation”



Finding “no relation” between variables
can also be statistically significant (and
clinically relevant).
If “n” is large then no relation probably
means that there really is no relation in the
general population (especially if “p” is very
small).
This can be just as clinically useful as data
in which there is a relationship.
Measuring the Magnitude
(strength) of relationships
between variables



In general the strength is a comparison of the observed
relationship to the maximum possible relationship between
variables (max possible that could have been observed).
If all WCC (white cell count) scores of males were equal
exactly to 100 and those of females equal to 102, then all
deviations from the grand mean (or normal) in our sample
would be entirely accounted for by gender. We would say that
in our sample, Gender is perfectly correlated with WCC, that
is, 100% of the observed differences between subjects
regarding their WCC is accounted for by their gender.
If WCC scores were in the range of 0-1000, the same
difference (of 2) between the average WCC of males and
females found in the study would account for such a small
part of the overall differentiation of scores that most likely it
would be considered negligible.
Explained Variation




Statistics explains the relationship
between variables.
Try to account for the relationship between
the variables that you are studying
compared to the overall differentiation in
the studied variable (dependent variable).
Like low HDL accounts for 10% of the total
variation in heart attack incidence or in the
above example gender accounts for x% of
white cell count.
That percent is the explained variation.
Calculating statistical
significance




This is “p” value.
Depends on the study and type of
analysis.
Always takes “n” into account as well as
the size of the relationship between the
variables.
Larger “n” means that a smaller
relationship can be significant.
Normal Distribution







This is the bell curve – it is assumed for most studies.
This is an assumed truth about the general nature of reality
and we use it as the basis for most statistical analysis.
It takes into account 2 features. Mean (or average) and
standard deviation.
IQ is perfect example. Mean = 100. Standard deviation = +
25. 68% of the population falls + 1 standard deviation from
normal and 95% falls + 2. Anything greater than + 2 is
“significant” (p = at least .05)
This is a calculated concept and the standard deviation is
calculated from your results.
The bigger the “n” the more closely the results will
approximate “normal distribution” even in something that isn't
well represented by a normal distribution.
There are ways to do statistical analysis without normal
distribution, but they aren't as powerful.
Null Hypothesis



All hypothesis are tested against the “null”
hypothesis. It is essentially a statement
that no relationships exist between your
tested variables. Ex:
Hypothesis = mice fed superfood will have
greater longevity then those fed normal
mouse food.
Null hypothesis = there is no relationship
between superfood diet and longevity.
The normality assumption



Keep in mind that we assume that all
variables ultimately meet the normal
distribution if the population studied is big
enough.
This is an assumption and can only be
proved in some cases – not all.
This may introduce some error, but we
honestly have no way to find/calculate that
error yet.